]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 466c191)
gguf-py : add support for endian conversion of BF16 data (#16594)
authorAleksei Nikiforov <redacted>
Wed, 15 Oct 2025 20:43:08 +0000 (22:43 +0200)
committerGitHub <redacted>
Wed, 15 Oct 2025 20:43:08 +0000 (22:43 +0200)
commit7adc79c03234de9a20661fd6dbf2d02c32ca7acb
tree8282f68f20c53b0f9c176d38c32070a692dc5530tree
parent466c1911ab736f0b7366127edee99f8ee5687417commit | diff
gguf-py : add support for endian conversion of BF16 data (#16594)

BF16 requires special handling in this script
while it's a 2-bytes data, but view is 1-byte by default.
Switch to correct view before attempting byteswapping.

With this change correctly byteswapping models like
Meta-Llama-3-8B-Instruct-bf16-GGUF
should be possible.
gguf-py/gguf/scripts/gguf_convert_endian.py diff | blob | history
Packaging of ggml-org/llama.cpp